Years ago I had a sat-nav that was absolutely certain about a shortcut. “Turn left,” it said. No hesitation, full of confidence. The road had been closed for two years.
It wasn’t lying. It just didn’t know.
I’ve been thinking about that little box a lot lately.
Because more and more, people are running their own SEO audits through AI — paste in the site, ask what’s wrong, get back a tidy list of fixes. And honestly? Some of that is genuinely useful. I’m not here to tell you the tools are rubbish. They’re not.
It’s all part of the same shift I wrote about when the bots quietly took over the web — AI’s doing more and more of the heavy lifting. Running your own audit is just the DIY end of it.
But there’s a catch, and it’s a sneaky one.
An AI audit feels powerful because it’s fluent, not because it’s right. It writes in full sentences. It sounds sure. And “sounds sure” is a very different thing from “is correct” — my sat-nav taught me that.
And when it can’t see something? It doesn’t go quiet. That’s the part that should worry you.
I had a client run exactly this recently. The crawler couldn’t get into the site at all — came back with a big fat nothing. Did the AI shrug and say “sorry, couldn’t read it”? Course not. It handed over a full, confident audit of a site it had never actually seen. Problems that didn’t exist. Fixes for pages that weren’t there. All delivered with the same certainty as everything else.
That’s the sat-nav cheerfully describing a road it’s never driven.
And that’s the dramatic version. The everyday version is quieter.
It only sees what it happens to look at. Ask what’s “missing” and it’ll cheerfully flag things sitting right there on the page — it just didn’t crawl that far. Now you’re off chasing a problem that was never a problem.
It has no memory. It doesn’t know you already tried that title change back in March and it did nothing. Every audit starts from a blank page, so you can get told to redo the very thing that already flopped.
It has no idea what industry you’re in. Ask it to “optimise the site” and you get the out-of-the-box answer — the same one it’d hand a dentist, a law firm, or a llama farm.
And the big one: if a standard prompt built your strategy, the business down the road can type the exact same prompt and get the exact same strategy. That’s not an advantage. That’s everyone turning up to the meeting in the same outfit.
Translation: the audit isn’t the valuable bit. It never was.
The valuable bit is knowing which of those “missing” flags actually matters and which is just the crawler tripping over an apostrophe. It’s remembering what we tried last quarter and why it didn’t land. It’s twenty-odd years of looking at this stuff and going, “yeah, ignore that one — but this one? This one’s worth your afternoon.”
AI is brilliant at producing an answer. It’s rubbish at knowing whether the answer is worth acting on. That part still needs a human who’s been round the block — ideally one who remembers where the roadworks are.
So by all means, run the audit. Just don’t hand your afternoon to a very confident box that’s never seen your street.
If you’ve been running your own AI audits and something feels a bit off — like you’re fixing things that were never broken — that’s usually the tell. Happy to take a look and sort the real flags from the noise.
Grab a slot: meetwithjames.ca. Bring the audit. I’ll bring the moderately cold coffee.
